49 replies to this topic

[WanWizard]

Cool, thanks for the feedback.

[Dream1975]

After I did an update today I saw some qt packages coming by (see attached). Unfortunately if I now start HbbTV it says HbbTV plugin for a second in the display and then the channel name again. It would seem the update broke it again unfortunately.

Attached Files

•  Upgrade.txt   11.77KB   10 downloads

[WanWizard]

Same race condition, packages are empty again.

 

This is a bug in the qtwebkit recipe, which assumes that data is always present,  which is not the case, that depends on whether or not other packages were build. And that leads to empty packages.

[Dream1975]

Same race condition, packages are empty again.

 

This is a bug in the qtwebkit recipe, which assumes that data is always present,  which is not the case, that depends on whether or not other packages were build. And that leads to empty packages.

Ok, so I have to wait for the next build i presume.

[twol]

Same race condition, packages are empty again.

 

This is a bug in the qtwebkit recipe, which assumes that data is always present,  which is not the case, that depends on whether or not other packages were build. And that leads to empty packages.

At the risk of being controversial , I only get this issue with the OpenPli build process (and fix it currently by forcing Qtstalker etc to be part of the image for the hd51), even though both  recipes-qt and meta-qt5 are largely the same as that in other images - so why do I never get an issue when building the other images?

[WanWizard]

If we know the reason, we knew the fix.

 

What usually happens is that recipe writers only take a single box / single build into consideration, usually also without rm_work.

 

In that case you don't have any issue with shared packages, arch issues, and files from other parts of the build process are always there. Another possible issue is parallel processing. If you write and build on a "low-end" box, this is not an issue. If you, like us, build with 16 parallel tasks, you can run into the issue that one task cleans up while another still needs the files. And this happens because the task dependencies are not defined in the recipe.

 

We have so many of these issues that by default we loop over every build command twice, just in case the first failed.

[twol]

You are correct in that I only run one receiver build at a time usually with a 4 core cpu, so that is a much simpler environment.

[WanWizard]

What it also can be, currently qtwebkit is build als an ARCH package, so it's shared between all boxes using that ARCH. Hoewever, if I look in the build tree, I see it uses MACHINE dependencies, such as the sysroot.

 

So it could well be the package compiles well for the first cortexa15 box, but when the second box starts, it doesn't compile (because it is already up to date), it is packaged from existing data, but that doesn't exist since the sysroot is MACHINE specific. So after that second run, all ipk's are empty.

 

I can find commits for qt_base and qtwebkit, where the ARCH was changed to MACHINE_ARCH.

./vs1500/qtwebkit_5.8.0+git0+74ac5b0f34-r0.1_vs1500.ipk
./vs1500/qtwebkit-mkspecs_5.8.0+git0+74ac5b0f34-r0.1_vs1500.ipk
./h9/qtbase-mkspecs_5.8.0+git0+49dc9aa409-r0.0_h9.ipk
./h9/qtbase_5.8.0+git0+49dc9aa409-r0.0_h9.ipk
./h9/qtwebkit-mkspecs_5.8.0+git0+74ac5b0f34-r0.0_h9.ipk
./h9/qtwebkit_5.8.0+git0+74ac5b0f34-r0.0_h9.ipk
./h9/qtbase-staticdev_5.8.0+git0+49dc9aa409-r0.0_h9.ipk
./h9/qtbase-plugins_5.8.0+git0+49dc9aa409-r0.0_h9.ipk
./h9/qtbase-tools_5.8.0+git0+49dc9aa409-r0.0_h9.ipk
./e4hd/qtwebkit-mkspecs_5.8.0+git0+74ac5b0f34-r0.0_e4hd.ipk
./e4hd/qtwebkit_5.8.0+git0+74ac5b0f34-r0.0_e4hd.ipk
./h7/qtwebkit_5.8.0+git0+74ac5b0f34-r0.1_h7.ipk
./h7/qtwebkit-mkspecs_5.8.0+git0+74ac5b0f34-r0.1_h7.ipk
./cortexa15hf-neon-vfpv4/qtdeclarative_5.8.0+git0+fffb997e19-r0.7_cortexa15hf-neon-vfpv4.ipk
./cortexa15hf-neon-vfpv4/qtxmlpatterns_5.8.0+git0+9f7e01b582-r0.7_cortexa15hf-neon-vfpv4.ipk
./cortexa15hf-neon-vfpv4/qtmultimedia_5.8.0+git0+544a8a4251-r0.7_cortexa15hf-neon-vfpv4.ipk
./cortexa15hf-neon-vfpv4/qtdeclarative-tools_5.8.0+git0+fffb997e19-r0.7_cortexa15hf-neon-vfpv4.ipk
./cortexa15hf-neon-vfpv4/qtsensors_5.8.0+git0+1220a31e6c-r0.7_cortexa15hf-neon-vfpv4.ipk
./cortexa15hf-neon-vfpv4/qtwebchannel_5.8.0+git0+e3ddde4b26-r0.7_cortexa15hf-neon-vfpv4.ipk
./cortexa15hf-neon-vfpv4/qtquickcontrols_5.8.0+git0+201c86391c-r0.7_cortexa15hf-neon-vfpv4.ipk
./cortexa15hf-neon-vfpv4/qtlocation_5.8.0+git0+41cc347040-r0.7_cortexa15hf-neon-vfpv4.ipk
./cortexa15hf-neon-vfpv4/qtwebsockets_5.8.0+git0+afb3c58d03-r0.7_cortexa15hf-neon-vfpv4.ipk
./hd51/qtwebkit-mkspecs_5.8.0+git0+74ac5b0f34-r0.1_hd51.ipk
./hd51/qtwebkit_5.8.0+git0+74ac5b0f34-r0.1_hd51.ipk
./hd51/qtwebkit-qmlplugins_5.8.0+git0+74ac5b0f34-r0.1_hd51.ipk

But for some reason, this doesn't happen for the other packages. So perhaps more bbappends are needed.

[ims]

today I tried reinstall qhbbtv on H7 because hbbtv does not work - problem with dependency ...

 

Tried in homebuild use bitbake -c cleanstate for qtbase, then build again qtbase, install qthbbtv again and all works well...

[WanWizard]

cleaning works, for one box and one build only. The build of the second box with the same ARCH will create empty ipk's again.